-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcalculate.py
188 lines (169 loc) · 7.97 KB
/
calculate.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
from multiprocessing import shared_memory, Queue, cpu_count
import numpy as np
import scipy as sp
from g_function import g_Function
from precompute import Precompute
import igl
from time import time
from random import randint
from workers import FaceWorker, EdgeWorker, RotationWorker
class Calculate:
"""
Calculates the deformation of a mesh.
"""
def __init__(self, V: np.array, F: np.array, precomputed: Precompute, g: g_Function):
"""
Initializes the calculation with a given mesh and precomputed data.
"""
self.V = V
self.F = F
self.precomputed = precomputed
self.g = g
self.random_seed = str(randint(0, 10000))
# Create shared memory for calculations.
e_ij_stars = np.array([V[self.precomputed.ev[i, 1]] - V[self.precomputed.ev[i, 0]]
for i in range(self.precomputed.ev.shape[0])])
nf_stars = igl.per_face_normals(V, self.F, np.array([1.0, 0.0, 0.0]))
u = np.zeros([self.precomputed.ev.shape[0], 3])
rotations = np.zeros([V.shape[0], 3, 3])
while True:
try:
self.shm_e_ij_stars = shared_memory.SharedMemory(
create=True, size=e_ij_stars.nbytes, name="e_ij_stars" + self.random_seed)
break
except FileExistsError:
self.random_seed = str(randint(0, 10000))
self.e_ij_stars_shared = np.ndarray(
e_ij_stars.shape, dtype=e_ij_stars.dtype, buffer=self.shm_e_ij_stars.buf)
self.shm_nf_stars = shared_memory.SharedMemory(
create=True, size=nf_stars.nbytes, name="nf_stars" + self.random_seed)
self.nf_stars_shared = np.ndarray(
nf_stars.shape, dtype=nf_stars.dtype, buffer=self.shm_nf_stars.buf)
self.shm_u = shared_memory.SharedMemory(
create=True, size=u.nbytes, name="u" + self.random_seed)
self.u_shared = np.ndarray(
u.shape, dtype=u.dtype, buffer=self.shm_u.buf)
self.shm_U = shared_memory.SharedMemory(
create=True, size=V.nbytes, name="U" + self.random_seed)
self.U_shared = np.ndarray(
V.shape, dtype=V.dtype, buffer=self.shm_U.buf)
self.shm_rotations = shared_memory.SharedMemory(
create=True, size=rotations.nbytes, name="rotations" + self.random_seed)
self.rotations_shared = np.ndarray(
rotations.shape, dtype=rotations.dtype, buffer=self.shm_rotations.buf)
# Create queues for workers.
self.face_work_queue = Queue()
self.face_complete_queue = Queue()
self.edge_work_queue = Queue()
self.edge_complete_queue = Queue()
self.rotation_work_queue = Queue()
self.rotation_complete_queue = Queue()
# Start workers.
self.face_workers = [FaceWorker(
self.precomputed, self.g, self.face_work_queue, self.face_complete_queue, self.random_seed) for _ in range(cpu_count())]
for worker in self.face_workers:
worker.start()
self.edge_workers = [EdgeWorker(
self.precomputed, self.g, self.edge_work_queue, self.edge_complete_queue, self.random_seed) for _ in range(cpu_count())]
for worker in self.edge_workers:
worker.start()
self.rotation_workers = [RotationWorker(
self.precomputed, self.g, self.rotation_work_queue, self.rotation_complete_queue, self.random_seed) for _ in range(cpu_count())]
for worker in self.rotation_workers:
worker.start()
def terminate(self):
"""
Terminates the calculation and cleans up shared memory.
"""
for _ in self.face_workers:
self.face_work_queue.put(None)
for _ in self.edge_workers:
self.edge_work_queue.put(None)
for _ in self.rotation_workers:
self.rotation_work_queue.put(None)
for worker in self.face_workers:
worker.join()
for worker in self.edge_workers:
worker.join()
for worker in self.rotation_workers:
worker.join()
self.shm_nf_stars.close()
self.shm_nf_stars.unlink()
self.shm_e_ij_stars.close()
self.shm_e_ij_stars.unlink()
self.shm_u.close()
self.shm_u.unlink()
self.shm_U.close()
self.shm_U.unlink()
self.shm_rotations.close()
self.shm_rotations.unlink()
def single_iteration(self, U: np.array, iterations: int):
"""
Performs a single iteration.
"""
# Initialization
e_ij_stars = np.array([U[self.precomputed.ev[i, 1]] - U[self.precomputed.ev[i, 0]]
for i in range(self.precomputed.ev.shape[0])])
nf_stars = igl.per_face_normals(U, self.F, np.array([1.0, 0.0, 0.0]))
u = np.zeros([self.precomputed.ev.shape[0], 3])
self.e_ij_stars_shared[:] = e_ij_stars[:]
self.nf_stars_shared[:] = nf_stars[:]
self.u_shared[:] = u[:]
self.U_shared[:] = U[:]
# ADMM optimization
for i in range(iterations):
# Update nf_stars
start = time()
for items in np.array_split(range(self.F.shape[0]), len(self.face_workers)):
self.face_work_queue.put(items)
for _ in range(len(self.face_workers)):
self.face_complete_queue.get()
print("Face time: ", time() - start)
start = time()
for items in np.array_split(range(self.precomputed.ev.shape[0]), len(self.edge_workers)):
self.edge_work_queue.put(items)
for _ in range(len(self.edge_workers)):
self.edge_complete_queue.get()
print("Edge time: ", time() - start)
self.e_ij_stars_shared = np.ndarray(
e_ij_stars.shape, dtype=e_ij_stars.dtype, buffer=self.shm_e_ij_stars.buf)
# Update u
for f in range(self.F.shape[0]):
for i in range(self.precomputed.fe[f].shape[0]):
e = self.precomputed.fe[f, i]
self.u_shared[f, i] += self.e_ij_stars_shared[e].dot(
self.nf_stars_shared[f])
# Calculate part of right hand side of the global step.
start = time()
E_target_edges_rhs = np.zeros([self.V.shape[0], 3])
for e in range(self.precomputed.ev.shape[0]):
v1 = self.precomputed.ev[e, 0]
v2 = self.precomputed.ev[e, 1]
w_ij = self.precomputed.L[v1, v2]
E_target_edges_rhs[v1, :] -= w_ij * self.e_ij_stars_shared[e]
E_target_edges_rhs[v2, :] += w_ij * self.e_ij_stars_shared[e]
print("E_target_edges_rhs time: ", time() - start)
# ARAP local step
start = time()
for items in np.array_split(range(U.shape[0]), len(self.rotation_workers)):
self.rotation_work_queue.put(items)
for _ in range(len(self.rotation_workers)):
self.rotation_complete_queue.get()
print("Rotation time: ", time() - start)
self.rotations_shared = np.ndarray(
self.rotations_shared.shape, dtype=self.rotations_shared.dtype, buffer=self.shm_rotations.buf)
# ARAP global step
start = time()
rotations_as_column = np.array([rot[j, i] for i in range(
3) for j in range(3) for rot in self.rotations_shared]).reshape(-1, 1)
arap_B_prod = self.precomputed.arap_rhs.dot(rotations_as_column)
known = np.array([0], dtype=int)
known_positions = igl.snap_points(
np.array([U[self.F[0, 0]]]), self.V)[2]
for dim in range(self.V.shape[1]):
B = (arap_B_prod[dim*self.V.shape[0]:(dim+1)*self.V.shape[0]] + self.g.lambda_value *
E_target_edges_rhs[:, dim].reshape(-1, 1)) / (1 + self.g.lambda_value)
new_U = igl.min_quad_with_fixed(
self.precomputed.L, B, known, np.array([known_positions[dim]]), sp.sparse.csr_matrix((0, 0)), np.array([]), False)
U[:, dim] = new_U[1]
print("ARAP time: ", time() - start)